VideoNet - Binary 3-Shot: leaderboard
Binary setting of VideoNet (Yadav et al., CVPR 2026) with in-context examples: before judging whether a test clip shows the named action, the model watches three other clips of that action. Accuracy in percent on the same balanced positive and hard-negative test clips as the zero-shot setting; higher is better.
Metric: Binary Accuracy (%; same binary test clips, three in-context example clips of the action). Source: arxiv.org. Saturation forecast: Around March 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (2026-03-05) (Medium) | 76.33 | #84 (GPT-5.4 (2026-03-05)) |
| 2 | GPT-5 (Medium) | 75.38 | #163 (GPT-5) |
| 3 | Gemini 3 Flash (Preview) (High) | 75.08 | #141 (Gemini 3 Flash (Preview)) |
| 4 | Gemini 3.1 Pro (Preview) (High) | 67.16 | #91 (Gemini 3.1 Pro (Preview)) |
| 5 | Qwen 3 VL 8B Instruct | 66.22 | #982 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=videonet-binary-3-shot · How It Works · Data refreshed daily, snapshot 2026-11-12.